import streamlit as st
import pandas as pd
from transformers import pipeline

# Load the dataset
@st.cache_data
def load_data():
    return pd.read_csv("insurance_data.csv")

data = load_data()

# Load DeepSeek model (General text classification)
@st.cache_resource
def load_nlp_model():
    return pipeline("text-classification", model="deepseek-ai/deepseek-llm-7b-chat")

classifier = load_nlp_model()

# Streamlit UI
st.title("Health Insurance Coverage Assistant")
user_input = st.text_input("Enter your query (e.g., coverage for diabetes, best plans, etc.)")

if user_input:
    # Detect intent using text classification
    result = classifier(user_input)  # Now we remove candidate_labels
    label = result[0]["label"]  # Get predicted label

    # Manual mapping (since DeepSeek does not support `candidate_labels`)
    if "coverage" in user_input.lower():
        intent = "coverage explanation"
    elif "recommend" in user_input.lower() or "best plan" in user_input.lower():
        intent = "plan recommendation"
    else:
        intent = "unknown"

    if intent == "coverage explanation":
        st.subheader("Coverage Details")
        condition_matches = data[data["Medical Condition"].str.contains(user_input, case=False, na=False)]
        if not condition_matches.empty:
            st.write(condition_matches)
        else:
            st.write("No specific coverage found for this condition.")
    
    elif intent == "plan recommendation":
        st.subheader("Recommended Plans")
        recommended_plans = data.sort_values(by=["Coverage (%)"], ascending=False).head(5)
        st.write(recommended_plans)
    
    else:
        st.write("Sorry, I couldn't understand your request. Please try again!")